Identification of two close homologs, dZip1 and dZip2, as specific zinc transporters involved in dietary zinc absorption

We previously demonstrated that dZnT1 is involved in the efflux of zinc from the midgut enterocytes for systemic use. However, it was not known which Zip is responsible for zinc uptake into the enterocytes. According to BLASTP searches for Drosophila homologs of mammalian Zip family members, the Drosophila genome encodes 10 putative Zip proteins (see Additional file 1: Figure S1A) [17]. Notably, the D. melanogaster genome lacks a close homolog of Zip4, a key player in mammalian absorption of dietary zinc. This was further confirmed when hZip4 and its closest D. melanogaster homolog CG10006 or foi were used as queries to blast across all genomes of various Drosophila species [40], suggesting that the role of Zip4 is executed by some other Zip homologs in the fly. To identify the Zip protein that mediates zinc uptake, we knocked down individually each of these putative zinc transporters, both ubiquitously (using daughterless- GAL4 or da-GAL4) and gut-specifically (using a gut-specific Gal4 line NP3084), and tested the sensitivity of the larvae to zinc depletion. A dramatic effect was observed with CG9430. When CG9430 expression was knocked down, either ubiquitously or gut-specifically, only around 10 to 15% of the larvae survived to adulthood on a zinc-limited diet (0.3 mmol/l EDTA-supplemented food) whereas the eclosion of the control flies was only slightly affected under the same conditions (Figure 1A; see Additional file 1: Figure S1C), suggesting that CG9430 is indispensable for proper zinc uptake.

The protein encoded by CG9430 shows high homology (28% identity and 50% similarity) to human Zip1 and Zip2 (hZip1 and hZip2). Interestingly, an immediately adjacent gene in this genomic region, CG9428, is very closely related to CG9430 (52% identity and 69% similarity) (see Additional file 1: Figure S1B). Compared with the CG9430 protein, the CG9428 protein displays slightly better homology to hZip1, and is in fact the closest homolog of hZip1 in the fly genome (29% identity and 48% similarity). We have therefore named CG9428 and CG9430 as Drosophila Zip1 (dZip1) and dZip2, respectively, hereafter. Both dZip1 and dZip2 are predicted to have the typical features of Zip family members, including eight transmembrane domains (TMDs), and extracellular amino and carboxyl termini.

The high similarity between dZip1 and dZip2 and their adjacent locations on the genome prompted us to further investigate whether dZip1 also participates in gut zinc uptake. Consistent with this notion, these two genes are the most highly expressed Zip genes in the gut, according to the fly atlas [41]. Because ubiquitous knockdown of dZip1 did not cause significant aberrance in viability, morphology, or fertility in flies fed on either normal or EDTA food, we decided to use a more sensitive assay to measure the effect on zinc absorption.

Activity of the secretory enzyme alkaline phosphatase (ALP) is very sensitive to zinc deficiency [42–46]. Indeed, ubiquitous or gut-specific RNA interference (RNAi) of either dZip1 or dZip2 significantly reduced the activity of ALP, but had no effect on the activity of the iron-dependent enzyme aconitase (Figure 1C,D) [47].

When we performed RNAi of both dZip1 and dZip2, and examined the additive effect when both genes were suppressed, virtually no larvae survived to adulthood, whereas a few larvae (< 10%) survived to adulthood when dZip2 alone was knocked down (Figure 1A; see Additional file 1: Figure S1C). Adding zinc back into the EDTA-containing food was able to restore the survival rates of dZip2-RNAi and dZip1, dZip2-RNAi flies to nearly normal level and to about 50% of normal level respectively, but other metals, including copper, manganese, and iron, did not have any ameliorating effect (Figure 1E). These results suggest that dZip1 and dZip2 are zinc-specific transporters, and while both are required for dietary zinc uptake, they are functionally partially redundant as well.

dZip1 and dZip2 are plasma membrane-resident zinc transporters responsible for zinc uptake into midgut enterocytes

During larval development, dZip1 is mainly expressed in the midgut, and is also present in trachea and testis, according to FlyAtlas Expression Data [41]. To examine the endogenous expression pattern of dZip1 at the protein level, we raised a polyclonal antibody against dZip1, and performed immunofluorescence staining on dissected larval gut.

Intensive expression of dZip1 can be detected in the midgut constriction (Figure 2A). Under higher magnification, the endogenous dZip1 was found to be localized to the plasma membrane of the enterocytes (Figure 2C), and interestingly, it was mainly restricted to the apical membrane (Figure 2B) of the enterocytes, which lines the lumen of the midgut. This apical localization is consistent with the role of dZip1 in dietary zinc uptake.

It has been reported that dZip2, when fused to enhanced green fluorescent protein (eGFP), presents a somewhat basolateral expression in the salivary glands [38]. We therefore investigated its location, particularly in the midgut. Because there is no dZip2 antibody currently available, we fused eGFP in frame to the C terminal of dZip2, and expressed the fusion protein in human Caco-2 cells. As predicted, dZip2-eGFP was found to be located on the plasma membrane of the Caco-2 cells (Figure 2D).

To examine whether dZip2 localizes to the apical side of midgut cells, we generated a dZip2-HA transgenic fly by fusing an hemagglutinin (HA) tag to the C terminal of dZip2, and expressed it in the midgut. A clear signal was observed on the apical membrane of the midgut (Figure 2E). These results thus suggest that both dZip1 and dZip2 mediate the absorption of dietary zinc from lumen into the cytosol of the enterocyte.

To confirm that dZip1 and dZip2 function as zinc importers, we monitored cytoplasmic zinc levels when they were overexpressed. We used the zinc-activating reporter MtnB-eYFP for this purpose. MtnB-eYFP comprises the regulatory sequence of the zinc-responsive gene metallothionein B (MtnB), an intracellular zinc binding protein, fused to an enhanced yellow fluorescent protein (eYFP) [31, 48, 49]. The MtnB-eYFP fluorescence signal was enhanced in dZip1-overexpressing larvae at the midgut constriction (Figure 3A), indicating excessive zinc accumulation in the cytosol of these cells. Consistent with the MtnB-eYFP fluorescence result, semi-quantitative reverse transcriptase (RT)-PCR also showed that both MtnB and MtnC were induced in flies with ubiquitous dZip1 overexpression (Figure 3B) [50], and these flies displayed specific sensitivity to dietary overload of zinc, but not to dietary overload of copper or iron (Figure 3E). Zinc accumulation, as detected by the zinc indicator Zinpyr-1, was also evident when dZip1 was expressed in Chinese hamster ovary (CHO) cells (Figure 3D). These observations confirmed that overexpression of dZip1 leads to zinc accumulation in the cytosol of cells.

Ubiquitous dZip2 overexpression leads to embryonic or first-instar larval lethality. This happened even when we used the gut-specific Gal4 driver. However, we identified a comparably weaker dZip2 line, in which ubiquitous expression still resulted in early larval lethality but gut-specific activation did not. This line, when gut-activated, exhibited zinc-specific sensitivity (Figure 3F).

We then used Caco-2 and CHO cells to study the zinc influx function of dZip2. Consistently, in dZip2-expressing cells, the MT2a transcriptional level, a reflection of cytoplasmic zinc level, was much higher than that of the control (Figure 3C). This zinc elevation was also evident when Zinpyr-1 was used as the zinc indicator (Figure 3D). These results indicate that dZip2 also transports zinc into the cytoplasm.

Intracellular zinc transporters are not significant in dietary zinc absorption

Our aforementioned experiments, along with previously published work [31] helped us to identify dZip1 and dZip2 as being required for transport of dietary zinc into the cytoplasm of the gut cells, and dZnT1 as being required to pump zinc out of the enterocytes into the hemolymph [31]. However, these zinc transporters are all plasma membrane zinc transporters, and whether the set of intracellular zinc transporters along the secretory pathway is involved in the zinc egress process is unknown. In dietary copper absorption, for example, the Golgi-resident Menkes gene ATP7A is critical and patients with Menkes disease exhibit severe bodily copper shortage [51–53]. If zinc absorption resembles that of copper, we would expect certain intracellular ZnT proteins, which mediate zinc efflux into the secretory pathway to be involved. The Drosophila genome encodes seven putative ZnT proteins, including five possible intracellular ZnTs, as indicated by homology comparison with mammalian ZnT proteins (see Additional file 2: Figure S2A) [17].

To analyze the functions of these ZnT proteins in gut zinc absorption, we collected all available RNAi lines from the Vienna Drosophila RNAi Center (VDRC), and custom-made another set at the Tsinghua Fly Center. ZnT35C was not further examined because it is not expressed in the midgut and was previously functionally characterized in malpighian tubules (FlyAtlas) [39]. Of the remaining ZnTs, CG31860 is not expressed in the midgut (FlyAtlas; see Additional file 2: Figure S2D); CG6672 (dZnT7) and CG5130 are analyzed in detail as shown in Figures 4 and 5; and the RNAi effect of the other two ZnTs, CG11163 and CG8632 based on RT-PCR analysis, are presented in Additional file 2 (Figure S2B,C). Most of these RNAi lines against intracellular zinc transporter genes caused lethality when ubiquitously activated (see Additional file 3: Table S1A), suggesting that these RNAi lines are working efficiently, and that the targeted ZnTs are indispensable for fly development.

However, when these intracellular zinc transporters were knocked down specifically in the gut by the gut GAL4 NP3084, none resulted in any defects in viability, morphology, or fertility, either under normal or zinc-deficient conditions (see Additional file 3: Table S1B), nor even at 29°C, a temperature at which the Gal4 protein is thought to be more potent and produces a stronger RNAi effect. To further scrutinize their involvement in dietary zinc absorption, RNAi of these intracellular transporters was carried out in a sensitized background with dZnT1 knockdown. When dZnT1 is knocked down, flies are more sensitive to zinc shortage (see Additional file 3: Table S1B) [31]. However, even under these conditions, the additional knock down (using gut-specific RNAi) of each of these intracellular ZnTs did not in any case noticeably increase the sensitivity of the dZnT1-RNAi flies to zinc deficiency (see Additional file 3: Table S1B).

As an independent and more sensitive test for bodily zinc deficiency, we quantified the ALP activity in the whole body minus the gut when these intracellular ZnTs were gut-specifically knocked down. We reasoned that a slight alteration in the zinc level might be reflected by a change in ALP, but might not result in overall survival or developmental phenotypes. We found that the ALP activity of all the gut-specific RNAi lines of intracellular ZnTs did not change (Figure 4F), indicating that intracellular zinc exocytosis is not significantly involved in dietary zinc absorption in the gut.

The Golgi apparatus is the harbor of the secretory pathway. ZnT7 has been shown to be important for the control of zinc levels in the Golgi apparatus [33–35, 42, 54]. A previous mouse Znt7 knockout study presented a perplexing scenario regarding the role of ZnT7 in dietary zinc absorption: while the Znt7-null mouse has a low overall zinc level, its tissues are not deficient in zinc. Furthermore, zinc supplementation cannot rescue the phenotype at all [33]. If indeed exocytosis is involved in dietary zinc absorption, we would expect zinc loading by the Drosophila ZnT7 counterpart to play a significant role.

Drosophila has only one likely ZnT7 homolog, CG6672. To confirm that this is indeed the Drosophila ZnT7, first we tried to determine the subcellular location of the CG6672 protein. To facilitate visualization, we fused eGFP in frame to the C terminal of CG6672. The Golgi marker was produced by fusing red fluorescent protein (RFP) behind the Golgi-targeted peptide of human β-1,4-galactosyltransferase [55–58]. We found that when CG6672 was co-transfected into Caco-2 cells, it apparently colocalized with the Golgi marker (Figure 4C), as described in a previous report [46]. Furthermore, the larval lethality resulting from ubiquitous silencing of CG6672 (Figure 4A) [17] could be partially rescued to late pupal or adult stage by expression of hZnT7, but not of dZnT1 (Figure 4A), corroborating that CG6672, hereafter named dZnT7, is the Drosophila ZnT7.

ALP activity is dependent on zinc loading in the Golgi. As expected, ubiquitous knockdown of dZnT7 significantly reduced zinc-dependent ALP activity, but had little effect on iron-dependent aconitase activity (Figure 4D). The affected flies exhibited a severe but hZnT7-rescuable phenotype: they died as late third-instar larvae. This experiment also showed the potent RNAi effect of the lines used. Consistent with the mouse knockout study, ubiquitous dZnT7 RNAi led to an overall reduction of zinc in the whole body (Figure 4E). To address precisely the function of dZnT7 in dietary zinc absorption, we tissue-specifically knocked down dZnT7 in the gut (using NP3084), and then examined the effects of this knockdown on the rest of the body. NP3084-driven dZnT7 RNAi did not appreciably affect the ALP activity of the whole body (Figure 4E). The above results suggest that dZnT7 in the gut does not contribute to systemic zinc levels, but rather that dZnT7 functions locally to regulate the activity of zinc-dependent enzymes. Therefore, zinc efflux to the Golgi does not seem to be significantly involved in dietary zinc absorption.

CG5130 (dZnT77C), a close homolog of dZnT1, also participates in the exit of zinc from enterocytes for systemic use

In the process of our screen for ZnT transporters involved in zinc absorption in the gut, the knockdown of CG5130, localized at 77C in the genomic region, was found to cause sensitivity in flies on EDTA-supplemented food when ubiquitously or gut-specifically knocked down (Figure 5E). In the phylogenetic tree of Drosophila ZnTs, CG5130 lies closest to dZnT1 (see Additional file 2: Figure S2A), sharing with it 26% identity and 48% similarity. Consistent with the plasma-membrane residence of dZnT1, CG5130 is also localized to the plasma membrane as shown by the fluorescence signal emitted by CG5130-eGFP in Caco-2 cells (Figure 5B). This is consistent with the previous finding showing the basolateral membrane localization of CG5130 in salivary glands [38]. CG5130 expression in CHO cells led to zinc reduction as indicated by Zinpyr-1 staining (Figure 5D).

To further address the zinc-absorption role of CG5130 in the gut, we examined the zinc-dependent ALP activity of the rest of the body when CG5130 was gut-specifically knocked down. Consistent with the observed sensitivity to zinc deficiency, gut-specific RNAi of CG5130 led to decreased ALP activity of the whole body minus gut, but did not significantly change aconitase activity (Figure 5C). Considering the sequence similarity between dZnT1 and CG5130, we investigated further, and found that this RNAi effect is not mediated by reduction of dZnT1, as CG5130 knockdown did not affect the expression of dZnT1 (Figure 5A).

We next investigated whether dZnT1 and CG5130 function cooperatively in the exit of zinc out of the gut into the circulation. We found that when the dZnT1-RNAi line was recombined with the CG5130-RNAi line, the double RNAi line displayed greater sensitivity to zinc deficiency compared with either of the individual single RNAi lines (Figure 5E). These data indicate that zinc efflux from the enterocytes is mediated by a collaborative function of dZnT1 and CG5130.

Expressions of the midgut zinc uptake genes are mainly influenced by changes in dietary zinc

One important question in the regulation of zinc absorption is how zinc transporters are regulated by available zinc. From the physiological point of view, the expression of zinc uptake proteins dZip1 and dZip2 should ideally be reduced when the diet is rich in zinc to avoid excessive zinc toxicity, and their expression should be increased when the diet is low in zinc, in order to facilitate dietary zinc uptake from zinc-limited food. Using semi-quantitative RT-PCR analysis (Figure 6), we found that this is indeed the case. Of all the Drosophila Zip genes, only dZip1 and dZip2 appear to be transcriptionally regulated by dietary zinc levels (Figure 6A). When the food was replete with zinc the RNA levels of these two zinc uptake proteins, dZip1 and dZip2, were repressed, and conversely, when the food was depleted in zinc, the RNA levels of dZip1 and dZip2 were increased. The induced dZip1 expression was further confirmed by immunostaining at the midgut constriction when dietary zinc was limited (Figure 6C).

These data suggest that dZip1 and dZip2 respond to dietary zinc availability, and function cooperatively to ensure appropriate zinc absorption under zinc-limited and zinc-supplemented conditions.

Assessing the transcription of all the ZnT genes in response to excessive dietary zinc, we detected a slight increase in expression only in the case of dZnT1 (Figure 6B). The immunohistochemical staining results suggest that most of this increase can be explained by de novo or ectopic induction of ZnT1 expression in regions other than the midgut constriction region [31]. This lack of obvious transcriptional control in the zinc absorption area (the midgut constriction region) prompted us to investigate whether dZnT1 is subject to post-transcriptional control. Quantitative analysis of dZnT1 in the midgut by western blotting demonstrated a strong post-transcriptional regulation of dZnT1; when zinc was high, the dZnT1 protein was dramatically reduced (Figure 6D).

We also assayed the expression of CG5130 by fusing it to an HA tag at the C terminal. However, there was no obvious change in the CG5130 protein level in response to zinc fluctuations (Figure 6E).

The midgut zinc uptake genes are unresponsive to bodily zinc status

Limiting zinc efflux when zinc is replete will benefit the rest of the body at the expense of the gut. This led us to investigate another important zinc regulation question: does zinc absorption reflect the bodily zinc requirement? To address this question, we genetically manipulated Drosophila to create these scenarios: high zinc in the rest of the body but low in the gut; high zinc in the gut but low in the body. We then determined how the influx and efflux zinc transporters responded. By introducing dZnT1 RNAi, we could make the fly zinc replete in the gut but zinc deficient in the rest of the body [31]. Under this scenario, dZip1 and dZip2 expression was much reduced (Figure 7A), despite the unsatisfied need for zinc in the body. By overexpressing dZnT1, we created the opposite scenario: higher zinc level in the body but lower in the gut [31]. In this case, dZip1 and dZip2 were both up-regulated (Figure 7A). Using dZip1 antibody, we can clearly see a much stronger signal of dZip1 when dZnT1 is over-expressed (Figure 7E). Another way to control body zinc level is through manipulating ZnT35C expression [39]. When ZnT35C is repressed or over-expressed, the body zinc level is correspondingly increased or decreased (Figure 7C) [39]. Again, we did not see any alteration of dZip1 and dZip2 expression (Figure 7B). Further, we did not see obvious dZnT1 expression change either (Figure 7D).

Taking these results together, we conclude that dietary zinc uptake is not responsive to the zinc status or need of the body. The regulation is, strictly speaking, controlled by the zinc status of the enterocytes, and not even directly by the diet itself. Dietary zinc influenced the expression of these uptake genes by affecting the zinc levels of the enterocytes.

Plasmids

UAS-dZip1 and UAS-dZip2 were generated by PCR amplification of the coding region of dZip1 and dZip2, respectively, from Drosophila cDNA, and cloned into vector pUAST [50]. Following this, pCDNA3.1-dZip2-eGFP, pCDNA3.1-CG6672-eGFP and pCDNA3.1-CG5130-eGFP were generated by fusing the enhanced green fluorescent protein (eGFP) in frame to the C terminal of the individual coding region of dZip2, C6672, and CG5130, respectively. All the constructs were verified by sequencing.

Cell culture and transient transfection

Human Caco-2 cells were maintained in DMEM medium (Invitrogen Corp., Carlsbad, CA, USA) containing 10% fetal bovine serum (FBS; Gibco BRL, Gaithersburg, MD, USA) at 37°C. Caco-2 cells were cultured and when they reached 80% confluence, they were transfected using Lipofectamine™ 2000 (Invitrogen) in accordance with the manufacturer’s instructions. A DNA: liposome ratio of 1.6 μg: 4 μl was used for each 4 cm² dish (0.8 μg DNA each when two plasmids were co-transfected).

Fly stocks, culture media, and transgenics

Fly stocks were raised at 18°C, and all the experiments were performed at 25°C on standard cornmeal food. When necessary, the food was supplemented with EDTA, or ZnCl₂ at the concentrations stated below for each experiment. The temperature was increased to 29°C when necessary. Fly stocks, including da-GAL4 and UAS-eGFP, were obtained from Bloomington Stock Center (Bloomington, IN, USA), while NP3084 and NP1093 lines were obtained from the Drosophila Genetic Resource Center at the Kyoto Institute of Technology (Kyoto, Japan). The RNAi lines were provided by the VDRC (Vienna, Austria) or custom-made at Tsinghua Fly Center (Beijing, China). Transgenic flies were prepared by P-element-mediated transformation in w ¹¹¹⁸ background.

Fly survival assays

da-GAL4 or NP3084 homozygous flies were crossed with transgenic flies, as indicated in each experiment. The progeny were fed on food supplemented with the stated metals or metal chelators. The density of each vial was about 70 progeny.

Antiserum preparation

A rabbit polyclonal anti-dZip1 antibody was raised against a synthetic peptide of amino acids 173 to 186 of dZip1 (DTEPQPHKDHHGHS) by Abmart (Shanghai, China).

Western blotting, immunohistochemistry, and microscopy

For western blot analysis, midgut samples were dissected, and homogenized in buffer containing 1% TritonX-100 in the presence of a proteinase inhibitor cocktail (P2714-1BTL’ Sigma Chemical Co., Chicago, IL, USA), centrifuged, separated by SDS-PAGE in 12% gels, and transferred to PVDF membranes (Millipore Corp., Billerica, MA, USA).

The antibodies used were anti-dZnT1, mouse monoclonal anti-β-actin, anti-HA, HRP-conjugated goat anti-mouse IgG, and HRP-conjugated goat anti-rabbit IgG (Zhongshan Goldenbridge Biotechnology, Beijing, China). Signals were developed with a SuperSignal West Dura kit (Thermo, Rockford, IL, USA). For immunohistochemistry, the anti-dZip1 antibody was preabsorbed with fixed w ¹¹¹⁸ embryos before being used for staining fly tissues. Third-instar larvae were dissected in cold phosphate-buffered saline (PBS), fixed with paraformaldehyde, and stained as previously described [31]. Anti-dZip1 and anti-HA were used (1:200 dilution) in combination with TRITC-conjugated goat anti-rabbit IgG (Zhongshan Goldenbridge Biotechnology Co. Ltd., Beijing, China). Samples were incubated in 50 ng/ml DAPI for 10 min for nuclear staining. Tissues were mounted in 50% glycerol/50%PBS. The fluorescence of MtnB-eYFP in the larval gut was captured with an ECLIPSE 80i microscope attached to a DXM1200F digital camera (both Nikon, Tokyo, Japan) The fluorescence of cell cultures was recorded by fluorescence microscopy (1X71; Olympus, Tokyo, Japan), and the immunostainings were examined using a laser scanning confocal microscope (FV500; Olympus, Tokyo, Japan).

Zinpyr-1 staining for intracellular labile Zn²⁺

dZip1, dZip2, and CG5130 were cloned into the pIRESneo vector, and transfected into CHO cells. Stable lines were obtained by selecting with 800 μg/ml G418. Cells were incubated in DMEM containing 75 μmol/l ZnCl₂ for 3 hours, washed 3 times with PBS for 5 minutes each time, and then fixed with paraformaldehyde for 30 minutes. After three washes, cells were incubated in 2 μmol/l Zinpyr-1 (Santa Cruz Biotechnology, Santa Cruz, CA, USA) in PBS for 30 minutes at 37°C. Excessive Zinpyr-1 was washed away by three washes in PBS, and the fluorescence signal was examined under an Olympus 1X71 fluorescence microscope.

RNA isolation and semi-quantitative RT-PCR

Total RNA was extracted from embryos, larvae, or gut from the third-instar larvae or from cultured cells, using TRIzol reagent (Invitrogen, Carlsbad, CA, USA). cDNA was transcribed from 1 μg total RNA with TransScript First-Strand cDNA Synthesis SuperMix (TransGen Biotech Co., Ltd., Beijing, China), in accordance with the manufacturer’s instructions. Semi-quantitative RT-PCR was performed using the specific primers corresponding to partial regions of the analyzed genes.

Measurement of ALP activity

Total protein (1 to 10 μg) was lysed in ALP lysis buffer (1.0 mmol/l Tris–HCl pH7.4, 0.5 mmol/l MgCl₂, and 0.1% Triton X-100), then 90 μl solution A (1.0 mol/l diethanolamine, 0.5 mmol/l MgCl₂ pH9.8) and 10 μl solution B (150 mmol/l p-nitrophenyl phosphate) were added. The absorbance at 405 nm was measured after incubation for 30 min at 25°C.

Measurement of aconitase activity

Total protein (10 to 100 μg) was incubated in PBS with Triton (137 mmol/l NaCl, 2.7 mmol/l KCl, 10 mmol/l Na₂HPO₄, 2 mmol/l KH₂PO₄, and 0.1% Triton X-100), and then mixed with 700 μl citrate reaction solution (30 mmol/l citric acid, 50 mmol/l K₂HPO₄ pH 7.4). The absorbance of the reactant at 240 nm was recorded for 2 min using a UV–Visible spectrophotometer, and the aconitase activity was calculated by the increased value of the absorbance.

Statistical analysis

Differences were analyzed by one-way ANOVA using SPSS software (version 16.0 for Windows; SPSS Inc., an IBM Company, Chicago, IL, USA). P < 0.05 was considered significant.